A systematic review of computational modeling of interpersonal dynamics in psychopathology

Interpersonal dynamics have long been acknowledged as instrumental to the generation and alleviation of psychopathology problems. Recent computational approaches have been argued to be uniquely suited for investigating such dynamics either with exploratory models (machine learning) or confirmatory models (generative modeling). However, the utility of such models to the study of interpersonal problems has not yet been scrutinized. We thus conducted a systematic review to assess the validity, reliability, and openness of computational models to the study of interpersonal psychopathology problems. Candidate studies (n = 2,944), including peer-reviewed conference manuscripts, were derived from five databases (MEDLINE, Embase, PsycINFO, Web of Science, Google Scholar) up to September 2024. A total of 55 studies met inclusion criteria and were assessed for their computational approach, validity, reliability, and openness. Results indicated that Bayesian modeling was the most commonly used approach (k=19), followed by machine learning (k=16), dynamical systems modeling (k=10), and reinforcement learning (k=10). Quality assessments revealed considerable heterogeneity in the validity of these approaches, with some scoring high primarily on empirical validity (reinforcement learning), others on theoretical validity (dynamical systems), and yet others on generative validity (Bayesian models). Finally, and most strikingly, few studies reported comprehensive performance metrics and even fewer studies adopted open science practices (specifically, 2 pre-registered their hypotheses and 8 shared their data and code online). We discuss these matters and conclude with more optimistic messages regarding how, when rigorously and openly conducted, computational approaches have the potential to advance the field of interpersonal psychopathology by enabling us to formalize and examine historically elusive social concepts (like mental representations of the self and others) and their role in psychopathology problems.